WorldTravel: A Realistic Multimodal Travel-Planning Benchmark with Tightly Coupled Constraints

Real-world autonomous planning requires coordinating tightly coupled constraints where a single decision dictates the feasibility of all subsequent actions. However, existing benchmarks predominantly feature loosely coupled constraints solvable through local greedy decisions and rely on idealized data, failing to capture constraint acquisition from realistic web interfaces. We introduce $\textbf{WorldTravel}$, a benchmark comprising 150 real-world travel scenarios across 5 cities, requiring agents to satisfy an average of 15+ interdependent temporal and logical constraints. To evaluate realistic deployment settings, we further develop $\textbf{WorldTravel-Webscape}$, a multi-modal environment with over 2,000 rendered webpages that preserve layout-dependent and information-dense travel interfaces, requiring agents to recover executable constraints from rendered web interfaces. Evaluating 10 frontier models reveals a severe performance collapse: GPT-5.2 achieves only 28.0\% feasibility in text-only settings, dropping to 3.4\% in multi-modal environments. We observe substantial degradation in planning feasibility when agents must recover executable constraints from rendered webpages, alongside a Planning Horizon threshold at approximately 10 constraints where reasoning reliability collapses. These findings suggest that realistic constraint acquisition and long-horizon planning remain complementary bottlenecks for current agents.

Many people now use AI assistants to help plan trips. But real travel planning is harder than it looks: choosing one museum ticket or dinner reservation can affect everything that comes after it. A missed time slot or a long queue may turn an entire itinerary into an impossible plan. We built WorldTravel to test whether today’s AI systems can handle this kind of realistic planning. Instead of giving models neatly organized information, we place them in a setting that resembles real travel websites, where they must read booking pages, compare options, and piece together scattered information across attractions, restaurants, and hotels. We found that even the strongest AI systems struggle. Models often fail to coordinate schedules or recover important details from web pages, especially when many decisions depend on one another. Our results suggest that AI still has difficulty turning information into reliable long-term plans, an ability needed for real-world assistants that help people organize complex tasks.